Melbourne University AES/MathWorks/NIH Seizure Prediction

Aindriú, I don’t think frequency is semantically monotonic either. In the time dimension, it is reasonable to think that an anomaly at 1:30 has the same meaning as another one at 6:45. However, IMHO it is not reasonable to think that an anomaly at bands 4,5,6 (however you compute them) has the same meaning as another one at bands 10,11,12. But the convolution operation in the TF plane is treating them as equivalent! Regarding your mean question, there’s a conv3D in Tensorflow. It’s meant for cross-correlation operations, so I’m not sure if it’s any useful.


Link to Full Article: Melbourne University AES/MathWorks/NIH Seizure Prediction